Thomas proposes that infants possess core capacities to identify and categorize relationships, and further suggests that such capacities can account for key findings in the study of early social cognition purportedly about other topics. Effectively, Thomas posits that infants are “pro-relationship” beings who (1) track and reason about the relationships between themselves and others as well as between third parties, and (2) prefer those who show evidence of initiating positive relationships with others. As part of this claim, Thomas argues that evidence suggestive of an innate moral core might be better explained by innate representations of social relationships: Rather than identifying who possesses positive traits such as helpfulness, infants simply prefer those who “initiate or maintain prosocial relationships” (Box 1). Though we do not dispute Thomas’ claim that infants show impressive capacities for relationship inference, we argue that the body of evidence to date falls short of suggesting that infants privilege relationship inferences over moral/trait-based ones.

First, Thomas reviews evidence that infants expect imitators to later help their imitated targets, but not novel targets; that is, infants make relationship-level inferences for the further prosocial actions of imitators, but apparently do not make trait-level inferences (Kudrnova, Spelke, & Thomas, Reference Kudrnova, Spelke and Thomas2024; Pepe & Powell, Reference Pepe and Powell2025). They imply that these relationship-specific tendencies likely extend to other kinds of prosocial actors beyond imitators (e.g., helpers, fair resource distributors, etc.), and thus infants’ apparently moral inferences are actually about relationships. We find this suggestion puzzling. For one, although evidence does suggest that infants fail to infer broad prosocial traits from a single act of imitation even at 18 months, 14-month-olds do apparently infer broad antisocial traits from a single morally relevant act (Gill & Sommerville, Reference Gill and Sommerville2023; Surian et al., Reference Surian, Ueno, Itakura and Meristo2018). We are also aware of no evidence that infants fail to make broader trait–level inferences about imitation (i.e., that an imitator of one target will not later imitate another); this evidence seems crucial to the claim that inferences about imitators are truly relationship-specific.

Second, Thomas’ central argument against an innate moral core requires that infants view imitation and helping as analogous (see Powell, Reference Powell2022; Spelke, Reference Spelke2022). Yet, in studies including both imitation and helping, infants readily distinguish those who help from those who imitate, and prefer those who help (Hamlin, Wynn, & Bloom, Reference Hamlin, Wynn and Bloom2007; Woo & Spelke, Reference Woo and Spelke2023); infants also prefer prosocial actors whose acts are not imitative (Geraci & Surian, Reference Geraci and Surian2011, Reference Geraci and Surian2023; Lucca, Pospisil, & Sommerville, Reference Lucca, Pospisil and Sommerville2018). Further, there is a well-documented “negativity bias” in early social cognition (see review in Vaish et al., Reference Vaish, Grossmann and Woodward2008); this is presumably the result of limited cognitive systems choosing to privilege avoiding harmful stimuli. Indeed, negativity biases are present throughout the early “moral core” literature: 3-month-olds distinguish hinderers from neutral characters, but not helpers from neutral (Hamlin, Wynn, & Bloom, Reference Hamlin, Wynn and Bloom2010), suggesting that negative social evaluation emerges first; relatedly, 6- and 10-month-olds expect a third party to avoid a hinderer but not to approach a helper (Chae & Song, Reference Chae and Song2018). Negativity biases are also observable neurologically: both 6- and 12-month-olds showed greater activity in the P400 ERP component (which selectively responds to social stimuli) after viewing hinderers versus helpers (Tan & Hamlin, Reference Tan and Hamlin2022; see replication in Soleimani, Reference Soleimani2023), suggestive of greater neural processing for hinderers. By contrast, we are aware of no empirical work within imitation demonstrating a negativity bias. Thus, although we agree that it would make sense to see imitators as better than neutral (they seem attentive to other agents’ actions and choose to reproduce them), it is unclear why humans’ cognitive system would view non-imitators – who in these studies merely produce a novel, neutral act that is irrelevant to infants themselves – as “bad” the same way they do hinderers. In sum, all of this work leads us to question whether the analogy between imitation and more morally relevant prosocial actions is appropriate.

Finally, Thomas posits that infants’ preference for prosocial over antisocial agents across diverse domains can be interpreted as being about relationships; that is, they desire to interact with those who initiate or maintain prosocial relationships. Although this pro-relationship view could account for infants’ basic preferences for prosocial agents, it is unclear how it explains findings from studies examining more complex questions. For instance, findings have shown that 4.5-, 8-, and 19-month-olds show context-dependent evaluations, preferring a character who hindered a previously antisocial target over one who helped it (Hamlin, Reference Hamlin2014; Hamlin et al., Reference Hamlin, Wynn, Bloom and Mahajan2011). If infants simply prefer agents who initiate prosocial relationships, they should uniformly prefer helpers regardless of the recipient’s previous acts, because helpers always initiate prosocial relationships. The same explanation similarly fails to explain findings in which 8-month-old infants themselves punish third-party antisocial actors via selective attention (Kanakogi et al., Reference Kanakogi, Miyazaki, Takahashi, Yamamoto, Kobayashi and Hiraki2022). If infants prioritize the establishment of prosocial relationships, they should avoid acting antisocially against third parties at all costs (as doing so is a sure way to establish an antisocial relationship). Further, a relationship-only framework struggles to explain why infants would selectively punish the aggressor instead of the victim, as punishing either agent could be interpreted as infants initiating a prosocial relationship with the other one. Instead, these findings seem better explained by more generic moral assessments: Infants view hinderers/aggressors as being “bad,” and thus it is seen as “good” to willingly engage in punishing them. This view is supported by evidence suggesting that by 13 months, infants associate antisocial actors with admonishment (DesChamps, Eason, & Sommerville, Reference DesChamps, Eason and Sommerville2016).

In summary, though Thomas’ proposition of a purely relationship-based model of the social world is intriguing, we believe it currently falls short of fully explaining findings in the sociomoral domain. Indeed, it may be premature to extend findings from work with imitation to other prosocial interactions, and key findings in the broader “moral core” literature are difficult to explain via relationship inferences alone. Thus, although Thomas convincingly argues that infants can and do make relationship-specific inferences and evaluations, we look forward to future work that further explores the circumstances under which infants make relationship-specific inferences versus more general, moral trait-based ones.